Anthropic has another big trick!
Historical moments: Claude's internal model conquered 18 years of rewarding speculation in the field of quantum computing.
In just three days, the undisclosed Claude model conquered the pending Aaronson-Ambainis conjecture since 2008!
Netizens speculate that the internal model appears to be an unpublished version of the Claude Model 4.
Lean's theories and theories have been formalized, and quantum computing's superiority to conventional computing has finally been answered.
In 1994, Peter Shor proposed the Scholl algorithm, known as the Earthquake.
It can break down large integers in polynomial time, and ordinary computers require exponential time to perform this task.
The foundation stone of the modern cryptography system has been shaken by this.
Scholl's algorithm blew the whistle, and then the term "quantum supremacy" swept through the global technology community and capital markets.
A bleeding blueprint seems to be right in front of us: quantum superposition, quantum entanglement, exponential acceleration.
It's as if by manipulating those fragile qubits, humans can launch a full-scale reduction strike, instantly entering an almost infinite new age of computing.
A blind cognitive illusion has begun to spread:
As long as there is quantum superposition, it seems that most problems, quantum computers can instantaneously go through all possible, achieving exponential attacks.
It's good that physicists and computer scientists soon realize:Scholl's algorithm is able to produce "index pressure" because of the strong cyclical structure behind the decomposition of integers.Quantum computers are good at capturing this repetitive law.
But if the input data is completely random, without any laws or special structures, can quantum computers achieve exponential transcendence?
In 2008, Scott Aaronson and Andris Ambainis, two leading scholars in the field of quantum computing, made abstraction a purely mathematical problem of physics and algorithms.
They likened the process of a computer reading input data to "observation".
In classical computing, each reader sees an exact "0" or "1"; quantum computers use superposition to distribute an observation over a mass state.
Two scholars have found that quantum computers produce "yes" probabilities after T-series observations, which can in essence be rigorously described as a "lower-order polynomial".
If this polynomial can prove that there must be a sufficiently large change in the inversion of a variable, i.e. a variable with significant influence, it can prove:Classical computers require only the extra cost of multiple levels to perfectly simulate the behavior of quantum computers.
That's the famous one.Aaronson-Ambainis speculates。
Now, this speculation has been proven by AI, and it could be called "the future of AI".Aaronson-AmbainisThe theory is this:
In their own words:
Quantum computing still has many unresolved puzzles, but now, there is one less.
It should be pointed out that quantum sovereignty has not disappeared, but this theorem kills the "blind superstition" of quantum computing and clarifies the "legitimate realm" of quantum advantage P.
quantum computing edition of the Oppenheimer momentPhysical Superintelligence isn't the first thing that's happened.
Last month, they solved five long-standing problems in quantum information theory.
Previously, Guy Blanc, Jordan Docter, Carmen Strassle and Li-Yang Tan of Stanford University have solved an example of this conjecture.
They've been reading the new evidence, thinking it's just a "big result":
It addresses a decades-old speculation as to which problems could bring about the essence of quantum acceleration.
This study covers several areas of study in the analysis of Boolean functions and probability combinations.
And this is not a proof of "learning the language", it is not a simple search of the literature that solves the problem, and it is a "beautiful way" that brings together all the content.
Perhaps the collaboration of experts in a number of human domains will ultimately prove this conjecture, but the speed of AI is hard to come by!
The first demonstration was completed three days after the completion of the Harness system, based on the Anthropic internal model.
He was immediately sent to be tested in a Lean-proving assistant.
But it's not over.
When model instances were asked to tackle unresolved problems using these intermediate results, they in turn refined these intermediate results, pushing the index from 22 to 11, then to 7, continuously downwards.
This is also a summary of the discovery of the process itself: the proof is not a flash of light, but a step closer to the optimal boundary in the repetitive grinding of the intermediate results.
Physics and mathematics are still vast, the universe is still shallow, and the unresolved mysteries that need to be explored by humans remain.
But one thing is certain:
The group of explorers who explored the boundaries of reason no longer stands alone and hangs in their shadows.
